Truly infinite worlds (poll)

[voxelproof]

Well, I think that there's a possibility for a really practically infinite worlds in Minetest. As far as I understand the principle of data storage for the MT worlds, their size is constrained by the capability of the data array used for that purpose. On the other hand, when a new world is started for the first time, no such data exist; the only given information that allows the map generator to create a new terrain is seed, algorithm and coordinates.

I suppose that at this point some may begin to see the direction of my reasoning. With probably not extremely difficult change of the map generating engines so that they allow the use of the long integer data type for coordinates it should be possible to generate terrain practically ad infinitum. And what about the database for all the interactions that a player performs within such an infinite world? Then it should be possible to use the command /set world which would set the center of a 60k x 60k x 60k cube where a player could build, destroy or play with other players. In the area beyond such actions would be impossible, however it still would remain open to the exploration.

And if a player found a more charming landscapes and decided that it would be a better place to settle down and make his base, then he could be able to set another old-style zone for his creations and fights, but that would erase all the data about his previous activities in the former area.

I think it's possible. The idea is partially inspired by the No Man's Sky's solution to that problem.

If such major improvement was made it would be a real game-changer for the Minetest. And the works of map generator contributors - Paramat and other developers, Gael de Sailly, Dokimi, Hume2 and others would at last get the chance to display their hidden potential in full.

Some topics that provided incentive for that proposal:

viewtopic.php?f=3&p=322341&sid=9da16b18 ... 9d#p322341
viewtopic.php?f=9&t=20066
viewtopic.php?f=9&t=20332

[Hume2]

Indeed, it would be a good feature. I was thinking that this could be done using a server modification. While the server would keep all so-far generated blocks in the infinite world, the client would see only a section with centre in his "home point". The home point would set to the location where the player would connect to the server. If the player tried to leave his section, his home point would be changed and he would be teleported accordingly. The player wouldn't notice anything except a change in coordinates and reloading the world. This would also hide the real player's coordinates.

When the world is big, there are many blocks those are never visited again or just a few. These blocks are usually not changed at all, the others are changed just a few. I think that there could be a timeout for these blocks. If the block was never changed, it could be forgotten to save space. If the block was changed just a few, it could slowly revert to its original state and then get forgotten. Maybe the ores shouldn't regenerate this way.

I think, it would be nice to have the world infinite. (Well, you have finite bits on your HDD which make finite combinations so the largest possible number stored on your HDD is also finite.) I noticed that the coordinates are stored in signed 16-bit integers which range from -32768 to 32767. It might look like it's enough for everything. However, there might be some kind of rare features those could be so rare so they would generate only a few times in this small world. And it would be sad if it didn't appear at all in a world.

[voxelproof]

Indeed, it would be a good feature. I was thinking that this could be done using a server modification. While the server would keep all so-far generated blocks in the infinite world, the client would see only a section with centre in his "home point". The home point would set to the location where the player would connect to the server. If the player tried to leave his section, his home point would be changed and he would be teleported accordingly. The player wouldn't notice anything except a change in coordinates and reloading the world. This would also hide the real player's coordinates.

When the world is big, there are many blocks those are never visited again or just a few. These blocks are usually not changed at all, the others are changed just a few. I think that there could be a timeout for these blocks. If the block was never changed, it could be forgotten to save space. If the block was changed just a few, it could slowly revert to its original state and then get forgotten. Maybe the ores shouldn't regenerate this way.

This is also interesting, however I'm not sure whether this would be as easy to implement as it seems to me that this idea of "nested" world is :) I think that small worlds as they are now also have some advantage - they are more "controllable" so to speak and certainly don't have the main open worlds' detering characteristic having sometimes powerful negative influence on players - the feeling of solitariness and even being lost in a vast and somewhat empty universe. But such a feature could be optional, as another type of map generators without making changes to the 'traditional' ones, as a testing ground for ambitious experiments with map generating algorithms rather than a new standard.

[rubenwardy]

Truly infinite worlds aren't possible (unless you like <1FPS)

[Dokimi]

A while ago I had an idea for this inspired by old sim-city maps.

Linked worlds.

When you get to the edge of the map, it shows a dialogue:
"You have reached the edge of this area, would you like to load the next area? Y N?"

Click yes, and it closes the current world, and opens a new one. The player is put in the corresponding position on the new map (e.g. if you dug down to the bottom, you would then go to the top of the new map).

Each map would have 6 links (N, S, E, W, top, bottom).

For simplicity the main menu would save one world name. It would recall which world the player was last in and load that.

No idea if this would work!

It would need special map gens too (for seamless transitions, all ground/space maps)

[843jdc]

All this sounds fine for a single player world. But it wouldn't be possible for multi-player mode with lots of players unless you had a *lot* of server hardware and the money to keep them running. Think Blizzard's online games.

Now server-hopping when you reached the end of one world that sends you to a server run by someone else would be cool. I'm not sure I would want a player automatically logging into my world from another world. Especially if the player is a server-hopping griefer.

There is a simulated world/game (that I can't remember the name of) where I read in the news that someone had paid a lot of real money to purchase a spot in that sim world. That game does allow world/server hopping. But the game is proprietary and seemed to require real money to purchase a spot on the giant world map to place the server.

There is an open-source version of that game. Though I didn't look at it for too long. It didn't seem similar to how Minetest is played.

[Hume2]

All this sounds fine for a single player world. But it wouldn't be possible for multi-player mode with lots of players unless you had a *lot* of server hardware and the money to keep them running. Think Blizzard's online games.

...

I think that deleting the never-edited parts might work for a server. And the slightly-modified parts could slowly revert to their original state and then get deleted. What do you think?

[Linuxdirk]

Truly infinite worlds aren't possible (unless you like <1FPS)

Infinite world != Infinite viewing range.

[Hume2]

Well, infinite. There's other problem. Your computer is too small to store graham number in its binary form and it's still a finite number. The limit must be somewhere because of the binary representation of coordinates. However, it might be still so big so travelling to the edge lasts so long so your life is too short to get there. (unless you teleport)

[Linuxdirk]

The limit must be somewhere because of the binary representation of coordinates.

The limit is the hardware.

[rubenwardy]

Truly infinite worlds aren't possible (unless you like <1FPS)

Infinite world != Infinite viewing range.

That wasn't what I was saying. Floats are limited in their precision, and having a bigger float doesn't make it unlimited. You'd need to have a number format which is capable of storing numbers approaching infinity, and for that you'd use some format with a carry on bit - ie: if the first bit is one, the next byte is part of the number. No GPU has support for doing maths on this data format. So you can't have a truly infinite continuous world and performance

You could use a string and a float position as a sort of region id and internal position, but this has a whole set of issues and difficulties with making it continuous whilst avoiding string comparisons whenever you use positions

The fact is that limitations need to be introduced in order to make optimisations.

If you relax the "truly" and make it "practically", then this is possible but it would still have a cost due to copying more data around

[Linuxdirk]

If you relax the "truly" and make it "practically", then this is possible but it would still have a cost due to copying more data around

Lets take the 64000^3 worlds we have now and map that to a coordinate system of 64000 steps in each direction. Now we can address 64000^3 sections of 64000^3 nodes (-214,218,7781:6658,-31014,7485 would be a valid node ID first defining the section and then defining the node).

This would cause a complete rewrite of large parts of everything but it absolutely is possible and would be practically unlimited and would allow hundreds of thousands of dimensions each hundreds of thousands of nodes high and wide

Oh, and maybe just wrap that in another "layer" so you can address the section, the sub section and the node, each of them having 64000^3 of the next-lower things in it. IDs will become pretty unreadable/unhandy (5874,-3147,-4310:-214,218,7781:6658,-31014,7485 would be a node ID - section_pos:subsection_pos:node_pos)

But to be honest I am pretty sure we'll stick with this rather small 64000^3 nodes worlds for a a couple of years or forever ...

[Hume2]

I think, we could use signed 32-bit integers instead of 16-bit integers. The world would be more than about 4 million km big. This takes a long time to cross and might be sufficient. If we used 64-bit integers, the world would be 1962 light years big which must be enough. Or is it still too small?

[Krock]

Hume2: there are a few issues with it:
1) Minutest uses 32-bit floating point for entity positions, which is very likely not precise enough.
2) It breaks the map format. It would need a conversion which would take a considerably long time on large maps
3) The potential of garbage increases enormously: Once generated and never reused mapblocks are never deleted
4) Effort to clean up above mentioned, unused mapblocks
5) How would you manage the coordinates e.g. for teleporting? Where was my house again? 932355235,24,-53263443 or 932356235,24,-53263443 ?
6) MInetest's code is not ready for this change. It would mean rewriting basically all position-related code to support larger sizes.

Some server admins already limit their map to 5 or 10 km to limit the map size and to keep the players together. Already now it's painful without teleporters to get anywhere distant if there's no railway or street nearby.
64 km ought to be enough for everyone.

[twoelk]

LOL, a poll to declare infinity mappable, yeah sure
I definitely see here a true understanding of the concept of infinity :-D
so now we just need a way to store numbers infinitely large,
need a way to store worlds that are infinitely large,
need infinite time to read and write infinite large data to infinite large databases,
of course infinite large power resources would come in handy for the infinitely large calculations needed
and of course we need players with infinite time to wait for a mere infinity for an infinite map to be processed by a server with infinite storage.

Even if we do replace the word infinite with just some version of "bigger than we will ever need" there are lots of technical issues we will run into. Allready now large maps become hard to handle and get buggier the bigger they are.

So please, please could we just drop this idea? 64 km ought to be enough for everyone.

probably not possible as everytime a past discussion gets forgotten somebody new brings up the discussion.
We might need a central place that explains the issues and explains the word "infinite" from a technical view that discussions like this can be pointed at.

Truly infinite worlds aren't possible (unless you like <1FPS)

probably wrong because a "truly" infinite world would in my opinion have also infinitely low FPS because infinity times whatever small amount still becomes infinity.

The only concept I really find possible would be to create some sort of grid of connected worlds each governed by some other server instance and stored in an individual database. The player-changing-worlds problem just needs to be solved in some for the player smooth way.

To understand the problem just imagine wanting number plates for cars that can store infinitely large numbers or old style paper based telephone books that can store infinitely large telephone numbers of an infinite amount of connections. The world we can grasp and process on our small planet simply is limited in the end although it might be larger than a minetest world.

[Linuxdirk]

"I think there is a world market for maybe five computers.", Thomas Watson, president of IBM, 1943

"Television won't be able to hold on to any market it captures after the first six months." - Darryl Zanuck, executive at 20th Century Fox, 1946

"There is no reason anyone would want a computer in their home." - Ken Olsen, founder of Digital Equipment Corporation, 1977

"Apple is already dead." - Nathan Myhrvold, former Microsoft CTO, 1997

"64 km ought to be enough for everyone." - Krock, Minetest developer, 2018

[Hume2]

Hume2, this idea was already mentioned in various topics but is finally not usable because
1) Objects can stand in between nodes, so we would need likely a 128 bit floating point number for such big maps (takes much computing time)
2) Breaks the map format. It would need a conversion from old to new (takes a very long time on large maps)
3) The potential of garbage increases enormously: Admins teleport to a far away location and the mapblock will never be used again
4) Effort to clean up above mentioned, unused mapblocks
5) Very painful to teleport far away. Where was my house again? 932355235,24,-53263443 or 932356235,24,-53263443 ?
6) Our code base is not made for this. It would mean rewriting basically all code with position values and Irrlicht itself if such enormous sizes are wanted.

Some server admins already limit their map to 5 or 10 km to limit the map size (garbage mapblocks) and to keep the players together. Already now it's painful without teleporters to get anywhere distant if there's no railway or street nearby.
So please, please could we just drop this idea? 64 km ought to be enough for everyone.

I agree with all these points. Indeed this would mean rewriting a large part of code. However, it's not definitely lost. It could be still done by server-side modding as I suggested above. If someone made a server with this feature, it might not be bad. Still, it would definitely require these things:
1) Rewriting a lot of code.
2) A way to delete garbage map blocks automatically.
3) Much more players to populate the world.

Mainly the third point is problematic. It would require at least thousands players if not millions. If anyone is willing to do this, it will be good but it's just a dream.

[Krock]

"64 km ought to be enough for everyone." - Krock, Minetest developer, 2018

Thanks for this input. I hope I'll be still on the list when you use it somewhen in the future again.

probably wrong because a "truly" infinite world would in my opinion have also infinitely low FPS because infinity times whatever small amount still becomes infinity.

Which again presupposes that everything is rendered. Surely, the larger the numbers get, the more must be calculated to keep the precision. Also if the world is split into smaller sectors which can be represented as whole numbers (idea from Linuxdirk), then it's still not possible to save an infinite large number without an infinite large disk. There's a reason why infinity is more of a concept than regular maths.
No Man's Sky is limited too, so if they claim that there are infinite many worlds, then they mean that in theory nobody could ever explore all of them.

[Linuxdirk]

"64 km ought to be enough for everyone." - Krock, Minetest developer, 2018

Thanks for this input. I hope I'll be still on the list when you use it somewhen in the future again.

I got the quotes from a random "tech predictions" list. Hopefully this could be used in future :)

Also if the world is split into smaller sectors which can be represented as whole numbers (idea from Linuxdirk), then it's still not possible to save an infinite large number without an infinite large disk.

Of course the limit is always the hardware. But 64k³ is simply too small. There is a German proverb "Haben ist besser als brauchen" (roughly translating to "having something is better than needing something").

Of course having 64k³ nodes is good and most use cases are covered, but you're still very limited. Having 64k³ times 64k³ sections of 64k³ nodes seems too much, but with this you don't have to think about space. Even if you don't need the space - you still have it (in case you might need it some time in the future).

Needing the space already is an issue when you want to have multiple dimensions (stacked instead of real other dimensions). With a:b:c (or only a:b) this won't be an issue at all because dimensions can be put in 64000,64000,64000.

[voxelproof]

Truly infinite worlds aren't possible (unless you like <1FPS)

Well, I was considering it for a while - does that mean that the sheer transition to the longer data type for coordinates would slow down the generators so much?

[voxelproof]

I think, we could use signed 32-bit integers instead of 16-bit integers. The world would be more than about 4 million km big. This takes a long time to cross and might be sufficient. If we used 64-bit integers, the world would be 1962 light years big which must be enough. Or is it still too small?

+1

I just tried to say that there should be a way to generate terrain on-the-fly with a decent speed without storing the generated data in the database. I'm deeply impressed by how that was achieved in Space Engine, the first and so far the only such realistic model of the Universe (even larger than No Man's Sky which, by the way is still regarded by many as very repetitive and falling short of expectations incited before the first release):







These screenshots which I made in SE are not posted to prove the superiority of this kind of graphics over the voxel blocky landscapes. As I figured out, within the voxel grid there's still a lot of place for improvement of landscape generation and, from the purely theoretical point of view it makes experimenting with landscapes much more easy and more fun.

[Dokimi]

Well, well, well...oh dear, what have I inspired!

Seeing as my mapgen, mg_tectonic, has inspired this discussion here's my two cents!

The real question: How big is big enough?

Okay, so I made mg_tectonic in a totally different style than the current default maps. Let's call mg_tectonic "realist style" - an attempt to realistically copy nature.

It does need big maps. But how big? Infinite?

The true limit is gameplay.

This is Realist style's actual major limit. Not the map limits per se. This has come up in this discussion too. The map becomes too big to be interesting. With the blocks and plants available it starts getting too hard to give enough small scale variation. Plus eventually the player wants to see something majorly different - unrealistically different.

We want to be able to climb a mountain that "feels" like Mt Everest, but we don't want to spend a realistic three weeks doing it.

I think mg_tectonic is already pushing the limits (at least with the current number of blocks/plants available). Realist style does encourage way more exploration - you do want to travel to the edge of the map, in all directions. But even then... I still haven't reached the map edge on mg_tectonic!

So... while infinite would be amazing, even doubling the map size would be more than enough to accommodate most realist style mapgens. So... if we want realist style maps... more plants and rock types would be equally important, perhaps more important, than bigger maps.

And that's my two cents!

[Hume2]

I noticed something similar in my terrainbrot. It could use a larger scale as well. The mountain ranges are currently too steep, the x/z axes could be scaled up at least four times.

However, even now it's hard to spawn in a world where both mountains and lowlands can be found in a small area. Usually I spawn on one island that contains mountains or lowlands but not both. And when I look for another island, it has usually the same landscape. Scaling the world up would mean that whole the world could consist of one type of landscape.

[Dokimi]

Here's something to put this discussion in context.

What would larger maps actually give someone making a realist style map?

[voxelproof]

The real question: How big is big enough?

I think that the horizontal size of Minecraft would do.

Okay, so I made mg_tectonic in a totally different style than the current default maps. Let's call mg_tectonic "realist style" - an attempt to realistically copy nature.

It does need big maps. But how big? Infinite?

The true limit is gameplay.

Yes, and this in turn depends on the type of gameplay. If someone is playing an exploration, the size of map which can be traversed in fly mode in less than 10 minutes is far too small, to say nothing of the capacity of such a map to host real terrain features. If there are to be realistically looking mountain ranges of the hight of more than, say, 3000 nodes, they simply won't fit into such a small world.

We want to be able to climb a mountain that "feels" like Mt Everest, but we don't want to spend a realistic three weeks doing it.[/qoute]

Speak for yourself ;)

I think mg_tectonic is already pushing the limits (at least with the current number of blocks/plants available). Realist style does encourage way more exploration - you do want to travel to the edge of the map, in all directions. But even then... I still haven't reached the map edge on mg_tectonic!

I once traversed the whole world along the road in lowerroad mod and it took just about 6 hours. Of course it would be boring if the textures and mapgen variation weren't very interesting.

So... if we want realist style maps... more plants and rock types would be equally important, perhaps more important, than bigger maps.

Maybe you're right. More diversity in the default content is certainly needed, but if someone says that increasing significantly the horizontal size of Minetest worlds is undoable I simply think that this must be related to some other constraints not related to the technicals, otherwise I'd suspect that this sort of assertment is utterly unprofessional.